One main challenge for statistical prediction with data from multiple sources is that not all the associated covariate data are available for many sampled subjects. Consequently, we need new statistical methodology to handle this type of “fragmentary data” that has become more and more popular in recent years. In this article, we propose a novel method based on the frequentist model averaging that fits some candidate models using all available covariate data. The weights in model averaging are selected by delete-one cross-validation based on the data from complete cases. The optimality of the selected weights is rigorously proved under some conditions. The finite sample performance of the proposed method is confirmed by simulation studies. An example for personal income prediction based on real data from a leading e-community of wealth management in China is also presented for illustration.
24 March 2016 in Journal of Nonparametric Statistics
Most feature screening methods for ultrahigh-dimensional classification explicitly or implicitly assume the covariates are continuous. However, in the practice, it is quite common that both categorical and continuous covariates appear in the data, and applicable feature screening method is very limited. To handle this non-trivial situation, we propose an entropy-based feature screening method, which is model free and provides a unified screening procedure for both categorical and continuous covariates. We establish the sure screening and ranking consistency properties of the proposed procedure. We investigate the finite sample performance of the proposed procedure by simulation studies and illustrate the method by a real data analysis.
Existing methods for handling nonignorable missing data rely on the correct specification of parametric models, which is difficult to check. By utilizing the information carried in an instrument, we propose a novel model selection criterion, called the penalized validation criterion, in the presence of nonignorable nonresponse with unspecified propensity. The proposed method can consistently select the most compact correct parametric model from a group of candidate models if one of the candidate models is correct. The population parameter estimators based on the selected model are consistent and asymptotically normal. Prior to our study, this type of result had been established only for ignorable missing data or nonignorable missing data under the strong assumption that the nonresponse propensity has a correct parametric model. Simulation results show that our proposed method works quite well. A real-data example is presented.
Aromatase inhibitors (AI) are frequently prescribed in gynecologic oncology. We sought to define the frequency and duration of AI use, characterize AI side effects and determine the reasons for discontinuation in these patients. Uterine and ovarian cancer patients with AI use for gynecologic cancer therapy were identified retrospectively. Data were abstracted from the electronic medical record, including cancer type, stage, prior cancer treatments, body mass index, concurrent medications, prevalence of AI side effects before and during AI therapy, length of AI treatment and reason for AI discontinuation. 146 women received AI therapy, with 68 for ovarian cancer (46.6%) and 78 for uterine cancer (53.4%). The majority (71.9%) had advanced stage disease at diagnosis. 54.1% noted AI-associated side effects within the first three visits after starting AI therapy. The most common side effects were arthralgias (29.5%), hot flashes (25.3%), new/worsening fatigue (16.4%), muscle or joint stiffness (8.2%) and myalgias (6.8%). The mean duration of therapy was 14.7 months. Gabapentin or selective serotonin reuptake inhibitor (SSRI) use was associated with decreased musculoskeletal side effects (gabapentin: p < .001, OR 0.88, 95% CI 0.83–0.94; SSRI: p < .001, OR 0.82, 95% CI 0.77–0.89). The most common reason for AI discontinuation was disease progression (87.9%), with 5.0% discontinuing due to side effects and 7.1% for other reasons. AI therapy for gynecologic cancers is frequently associated with musculoskeletal side effects, but rarely leads to treatment discontinuation. Thus, AI side effects should be assessed in gynecologic cancer patients to allow potential mitigation of symptoms through adjunct therapies.
Theoretical results of frequentist model averaging mainly focus on asymptotic optimality and asymptotic distribution of the model averaging estimator. However, even for basic least squares model averaging, many theoretical problems have not been well addressed yet. This article discusses asymptotic properties of a class of least squares model averaging methods with nested candidate models that includes the Mallows model averaging (MMA) of Hansen (2007, Econometrica 75, 1175–1189) as a special case. Two scenarios are considered: (i) all candidate models are under-fitted; and (ii) the true model is included in the candidate models. We find that in the first scenario, the least squares model averaging method asymptotically assigns weight one to the largest candidate model and the resulting model averaging estimator is asymptotically normal. In the second scenario with a slightly special weight space, if the penalty factor in the weight selection criterion is diverging with certain order, the model averaging estimator is asymptotically optimal by putting weight one to the true model. However, MMA with fixed model dimensions is not asymptotically optimal since it puts nonnegligible weights to over-fitted models. The theoretical results are clearly summarized with their restrictions, and some critical implications are discussed. Monte Carlo simulations confirm our theoretical results.
13 June 2020 in Annals of the Institute of Statistical Mathematics